2013 IEEE Symposium on Computers & Informatics
Enhancement of Iris Boundary Detection based on Otsu Method Ali Abd Almisreb1 & Nooritawati Md Tahir2 Faculty of Computer Science & Mathematics 2 Faculty of Electrical Engineering Universiti Teknologi MARA 40450 Shah Alam Selangor Malaysia Corresponding Author:
[email protected] 1
Abstract- The aim of this paper is to propose a new method for pupil boundary detection. Iris recognition system is considered as one of the most dependable identification system. Anyhow, these systems are still not widespread due to many weaknesses, for example, the processing time is still not instant that lead to ineffectual identification method. One of the major tasks in iris recognition systems is accurate detection of the iris. However, pupil boundary detection method depended on the circular or elliptic shape of the iris which could lead to inaccurate results of boundary detection. Thus, this can affect on the next step of iris recognition systems. Hence this research proposed thresholding the eye image using Otsu’s methods to remove the pupil. Our method was tested on all CASIA V3.0 eye images and 100% accuracy of iris boundary detection is attained with acceptable computational time that suitable to be implemented for real time processing. Keywords: Iris recognition, Iris boundary detection, Otsu’s method
I. INTRODUCTION A great attention has rapidly gained by biometrics systems in the past few years. Biometrics technologies have been applied in various fields such as a passport checking, border control, access doors control, banking services. The highest recognition accuracy rate is achievable by iris recognition system as compared to all other biometrics systems such as fingerprint, face, voice, hand geometry, and ear recognition. In iris recognition concepts, iris segmentation is deemed to be the most important step in any iris-based recognition system to accompany pupil boundary detection as the first vital phase to detect the required iris. For instance, in 2004, Daugman [1] has suggested a robust approach named integro-differential operator to extract both of pupil-iris and iris-sclera boundaries. Another sufficient and famous method to detect the pupil boundary was addressed by Wildes [2] where he analyzed the eye image edges followed by circular Hough transform to detect the boundaries. On the other hand, Kumar et al. [3] suggested a scheme based on correlation filters for iris recognition. Firstly, the iris detection method proposed to down-samples an iris images to
978-1-4799-0210-1/13/$31.00 ©2013 IEEE
100 by 100 pixels. After that, it calculated the cross correlation between the coarse image and a bank of 100 circles of variant radii. Then, a gradient operator is passed to the resulting circular integrations to find the maximum radial gradients, a factor for the pupil-iris boundary and another factor for the iris-sclera boundary. Many researchers have worked to raise the localization accuracy of the eye pupil. This can be noticed in some of the researches that have done by Jarjes etal [4] that localized the pupil using a circle which is fitted to the set of pupil boundary pixels detection by using the angular integral projection method. Then, the accurate contour is addressed using active contour model initialized on the pupil circle. On the other side, Kallel [5] used a morphological cleaning approach to pre-process the pupil binary image, where by applying this method prior integro-differential operator made iris segmentation faster by 18 times. Further, Zhang [6] segmented the pupil boundary gradient image of iris directly without need of filtering and specific normalization function is accomplished. Hence in this paper, we will deem further to investigate the capability of Otsu method for detection of iris boundary. II.
LITERTATURE REVIEW
There are numerous approaches which are being applied for iris recognition purpose. Generally, the first step in iris recognition system is capturing the human’s eye image followed by localize the iris part from the rest of the image, after that normalized this segmented region and finally will be the feature extraction process to get an iris template which represents the most related features to the acquired eye image. Performance of all subsequent phases in an iris recognition system greatly relies on accuracy detection of pupil-iris and iris-sclera boundaries in the eye images. Earlier, Ziauddin and Dailey [7] has implemented one such system in which pupil boundary is determined using image gray levels as well as utilise edge detection and circular Hough transform to localize iris boundary. A number of optimizations are introduced to traditional Hough transform based methods. Further, Mohammed et al. [8] suggested an efficient iris localization method based on the angular integral projection
143
2013 IEEE Symposium on Computers & Informatics
function (AIPF) to detect the iris boundaries in iris images. The addressed algorithm adopted boundary pixels detection and curve fitting. First, the estimated pupil centre is calculated. Then, two sets of radial boundary coordinates are obtained for the iris inner and outer boundaries using AIPF algorithm. Lastly, the iris boundaries by adapting a circle for each of the pervious boundary pixels are set. In the recognition stage, 2D Gabor filter is used to extract the iris code for the normalized iris image. At minimum, the suggested method was experimented on CASIA V1.0 iris images database based on performance measure of both False Accept Rate (FAR) and False Reject Rate (FRR) curves. Experimental results showed that the proposed iris localization algorithm is indeed efficient and improved iris recognition. Next, Koh et al. [9] has addressed a suitable iris detection algorithm that applied an active contour model and a circular Hough transform is implemented. Experimental results on 100 images from CASIA iris image database showed that this method achieved 99% accuracy and is about 2.5 times faster than the Daugman’s in locating the papillary and the limbic boundaries. But Li et al. [10], with the aim to enhance the efficiency of the iris detection, a novel method is suggested which combined the features of Hough transform and wavelet analysis. For the inner edge detection, the wavelet analysis is implemented with the iris image to amputate the noise section. Then applying Hough transform to localize the iris inner edge precisely. Thereby ignore all the results that contained noise. For detect the outer edge, the data of iris inner detection are used to remove noise so as to get effective edge information. Based on this approach, an enhanced Hough transform is applied to represent the iris outer edge precise and fast.
III. METHODOLOGY In this section the proposed method will be discuss in detail. a. Pre-preprocessing Pre-processing stage is to clean up and binarize the eye images, the resulting image will be more contrast than the original image. Otsu’s method will be used to achieve this step. The method supposes that the image to be thresholded contains two parts of points such as foreground and background. Then computes the best threshold that divides those two regions so their combined spread is minimal. The probability of occurrence of gray level i is defined as:
In the bi-level thresholding method, the pixels of image are divided into two classes C1 with gray levels [0,1,…,t] and C2 with gray levels [t+1,…L-1] by the threshold t . The gray level probability distributions for the two classes are:
ୀ ିଵ
ܹଶ ൌ ܲݎሺܥଶ ሻ ൌ ܲ ୀ௧ାଵ
b. Illumination Removing To remove the illumination that appears in the pupil center we used Flood Fill method to overcome this problem. Here is a pseudo code that explained the main points in this method. Flood-fill (N, X, Y): (A) • If ( color(N) X, Go to (A). • Else • Color (N) = X. o Flood-fill (N,X,Y-1). o Flood-fill (N,X,Y+1). o Flood-fill (N,X-1,Y). o Flood-fill (N,X+1,Y). • Go to (A). c. Pupil Segmentation Finding the eye iris is not an easy task as its intensity is similar to that of the sclera and is often covered by eyelashes and eyelids. However, for the reason of its regular size and uniform dark shade of the pupil, is considered easy to segment. Both pupil and iris can be estimated as concentric and this provides an accurate entry pixel for auto localization. Enrichment median filter is implemented on the raw eye image for the reason to eliminate noise from the image and to support the clustering in the pixel histogram. The outcome histogram is built to identify a point that can be used to threshold the image. The pseudo code for the filter is as outlined below: allocate output Pixel Value; edgex := (window width / 2); edgey := (window height / 2); for x from edgex to image width - edgex for y from edgey to image height - edgey allocate colorArray for fx from 0 to window width for fy from 0 to window height colorArray[fx][fy] := input Pixel Value [x + fx - edgex][y + fy - edgey] sort all entries in color Array output Pixel Value [x][y] := color Array [window width / 2][window height / 2]
ିଵ
݊ ܲ ൌ ǡ ܲ Ͳǡ ܲ ൌ ͳ ܰ
ݓ ൌ
௧ೕ
௧
ܹଵ ൌ ܲ ݎሺܥଵ ሻ ൌ ܲ
ܲ
ୀ௧ೕషభାଵ
144
2013 IEEE Symposium on Computers & Informatics
Next step is to apply the zero cross edge detector to mark the pupil boundaries. Laplacian of Gaussian filter is considered core of the zero crossing detector. As described there, `edges' in images give rise to zero crossings in the LoG output.
IV EXPERIMENTAL AND RESULTS In this section the algorithm proposed is evaluated using CASIA Iris V3.0 as the database. This database comprised of 22,035 iris images from more than 700 subjects. All iris images which used in here are 8 bit gray-level JPEG files, captured under near infrared lightening. A sample of the iris image is as shown in Figure 1.
Figure 3: Removing illumination by Flood Fill Algorithm
After removing the illumination, the pupil boundary can be detected. Firstly, median filter will be utilized repeatedly as in Figure 3, in order to increase the regions contrast as depicted in Figure 4.
Figure 1
Firstly, in order to detect the pupil boundary, the image is threshold using Otsu’s method and the output after reflecting the image is as shown in Figure 2.
Figure 4: Applying filter The next step is to draw the pupil boundaries for this objective using zero cross filter to binarize and detect the pupil boundary as shown in Figure 5.
Figure 2: After reflecting Otsu's method
The raising problem here is the illumination. Hence, to remove it the Flood Fill is used to combat the two areas inside the pupil region as shown in Figure 3.
Figure 5: Pupil boundary by zero crossing
Finally, the last final step is to combine image in Figure 5 with original image as in Figure 1 for detection the final pupil boundaries as illustrated in Figure 6.
145
2013 IEEE Symposium on Computers & Informatics
REFERENCES
Figure 6: Detection of pupil boundary With CASIA IRIS V3.0 as database, perfect accuracy of iris boundary detection is attained based on our proposed method with computational time of 0.3 seconds. IV.
CONCLUSION
In conclusion, an efficient and accurate approach for iris boundary detection is proposed, regardless of the iris pupil shape. This will guide to a more dependable and sturdy iris recognition system. In addition, based on the Otsu’s method for thresholding and technique for pupil and iris detection, it was also proven that the proposed method is capable of detecting the pupil boundary perfectly 100%. The obtained average computational time is reflected the appropriateness of this approach to be used for real time iris recognition application.
[1] J. G. Daugman, "How Iris Recognition Works," IEEE Transaction on Circuits and Systems for Video Technology, 14 (I), 21-30, 2004. [2] R. P. Wildes, "Iris recognition: An Emerging Biometric Technology,” Proceedings of the IEEE, 85 (9), 1348-1363, 1997. [3] B. Kumar, C. Xie, and J. Thornton, “Iris Verification Using Correlation Filters” in Proc. 4th Int. Conf. Audio and Video-Based Biometric Person Authentication, 697–705, 2003. [4] Jarjes, A.A.; Kuanquan Wang; Mohammed, G.J.; "Iris localization: Detecting Accurate Pupil Contour And Localizing Limbus Boundary," Informatics in Control, Automation and Robotics (CAR), 2010 2nd International Asia Conference on, 349-352, 2010. [5] Kallel, I.K.; Masmoudi, D.S.; Derbel, N., "Fast Pupil Location for Better Iris Detection," Systems, Signals and Devices, 2009. SSD '09. 6th International Multi- Conference on, 1-6, 2009. [6] Zhang Zhijia; Zhang Haifeng; Yuan Weiqi; "An Iris Location Method Based on the Active Contour," Computer Science and Information Engineering, 2009 WRI World Congress on, 537-541, 2009. [7] Sheikh Ziauddin; Matthew N. Dailey,” A Robust Hybrid Iris Localization Technique”, Computer Science and Information Management Asian Institute of Technology, 2009. [8] Ghassan J. Mohammed, Hong BinRong, and Ann A. Al-Kazzaz,” A New Localization Method for Iris Recognition Based on Angular Integral Projection Function” School of Computer Science and Technology Harbin Institute of Technology,2009. [9] Jaehan Koh, Venu Govindaraju, and Vipin Chaudhary,” A Robust Iris Localization Method Using an ActiveContour Model and Hough Transform” Department of Computer Science and Engineering, University at Buffalo (SUNY), 2010. [10] Jing-Hui Li; Jian-Hua Wu; De-Xuan Zou, “New Algorithm of Iris Localization” 2009 World Congress on Computer Science and Information Engineering, 2009. [11] Almisreb, A.A; Tahir, N.M.; Ismail, A.I.; Abdullah, R; “Enhancement pupil isolation method in iris recognition”, Proc of IEEE International Conference on System Engineering and Technology (ICSET), 2011.
ACKNOWLEDGMENT This research work was funded by Universiti Teknologi MARA RIF Grant No: 600-RMI/DANA 5/3/RIF (305/2012).
146